Lessons learned from the co-development of operational climate forecast services for vineyards management

This study describes the process of co-developing an operational climate forecast service for the viticulture sector. Weather and climate conditions affect grapevine development in cultivars: anticipating the atmospheric variables in the coming weeks and months is thus relevant for effectively managing vineyards, as impacts will be felt in wine production, biodiversity, and a wide range of related aspects. The operational service was co-developed with two types of users: impact modellers, who are the intermediary users incorporating climate forecast outputs in phenological and disease models, and end-users from the wine sector with vineyards in various European locations. For the operational service, sub-seasonal and seasonal climate forecasts were tailored considering their needs. The initial steps of the co-production process identified relevant decisions for which the service was essential and co-defined effective ways to deliver the climate information. Afterwards the climate forecasts outputs were integrated with impact model data. Substantial efforts were directed at the harmonisation of climate services information with the decision-making system of end-users. Because end-users need to navigate, comprehend, and select from various alternative options amidst uncertainty, significant emphasis has been placed on crafting the visual representation of the climate service, incorporating interactive elements, and cognitive considerations, thereby enhancing the overall user experience. Practical implications: Weather and climate conditions affect the development of cultivars: phenological stages, disease risk, or wine quality. This study presents the co-production process for deploying a climate service for vineyard management. The service consists of the near-real-time operational provision of sub-seasonal and seasonal climate forecasts. This service aims to help vineyard managers make decisions by anticipating the climate conditions up to three months in advance for a set of essential climate variables. The service includes the requirements of a set of intermediary users, which are disease risk and phenology researchers who feed their impact models with the provided climate forecasts. To our knowledge, it is the first time that sub-seasonal and seasonal climate forecasts have been integrated into an operational service for vineyard management.The service was developed during the VitiGEOSS project, the primary outcome of which is the VitiGEOSS platform: a single entry-point solution for wine producers aiming to boost vineyard sustainability. Three pilot plots are used for the service development: the Douro region in Portugal, the Catalonia region in Spain, and the Campania region in Italy. The co-production is seen as an iterative, interactive and collaborative process that brings together a plurality of knowledge sources to mutually define problems and develop usable products to address these problems. The process can be divided into three phases: initial co-exploration, co-design and co-development.The co-exploration phase consisted of conducting a benchmark analysis, identifying available services to viticulture and highlighting the gap that the newly developed solution could fill in. In the co-design phase, more intense engagement methods were implemented to communicate the capabilities of climate forecasts and design the product’s visualisation. Finally, the co-evaluation consisted of an ad hoc and ongoing process during the recurring meetings and a final workshop conducted with direct and potential users to obtain feedback on using the platform and potential further developments.As a result, the operational post-processing workflow of the sub-seasonal and seasonal state-of-the-art model outputs was adapted to include users’ requirements. The workflow includes the downscaling of model outputs to increase their spatial resolution, the storage of sub-seasonal climate forecasts at daily frequencies to feed phenological impact models, the provision of past climate simulations to train the impact models, the bias-adjust of climate forecasts to reduce systematic model errors, the assessment of the climate forecasts to aware users on their quality and the deployment of a server to allow access to impact modellers.The variables provided are mean, minimum and maximum temperature, accumulated precipitation and solar radiation. The final co-designed product consists of a graphic of bars. The user can select the location and variable, and the forecasts for the next four weeks or the next three months are displayed for the sub-seasonal and seasonal climate forecasts, respectively. For each week or month, three bars show the tercile probabilities of above-normal, normal or below-normal conditions for the time of the year. The limits of the tercile categories are also displayed, which helps to understand the averaged values of the variable in the selected region and the time of the year calculated with information from the past. The probabilities of extremes, i.e. exceeding (non-exceeding) the percentile 90th (10th), are displayed as extra bars for each week and month in case the prediction is skilful and they are greater than 40%.Last but not least, the results of the skill assessment are shown. This skill assessment consists of calculating a metric for each location, initialisation date and forecast time. When this metric, known as skill score, is positive (negative), the climate model forecast is better (worse) than the climatology (the climatology is a naive forecast which considers all categories equiprobable by definition). If the skill is negative, which means there is no proof of the added value of using the numerical model prediction, the bars are blurred, and the user can still be informed of the tercile category limits.The service has been provided for almost two consecutive years. During this period, we have confirmed the value of some key recommendations from literature when co-producing a climate service. Especially relevant is the need to invest time and resources to conduct a good co-production process by repeating the iterative process. We also tested the idea of providing information in the format of seasonal climate forecast outlooks to end-users before the service was operational. In this way, the climate service process would be ‘slowed down’, by which the use of climate data is incorporated more slowly and allows for place-based knowledge from the resource managers to be included too.One of the most challenging questions is the communication of the probabilistic forecasts product and their quality. The probabilistic nature of climate forecasts is sometimes understood by novice users as the incapacity to provide a deterministic forecast. Similarly, negative skill leads users to interpret the forecast fails. Overall, the perception of the service quality can be low. To avoid this perception, the co-producing process is essential to engage with users and redefine the service if possible.In this work, we have tried different ways to overcome low-quality perception. The users requested access to the climate forecasts for their vineyard plots. For that reason, the downscaling technique was applied to increase the spatial resolution, and the bar plots were designed to show the forecast for a specific location. However, to improve the perception of the forecast quality, the forecast can be contextualised by providing the forecast in the format of a map of the surrounding areas.Despite all these limitations, end-users reported on the extreme importance of climate information. Furthermore, end-users can benefit from getting the climate services integrated with the impact models to take maximum advantage of all services and make timely decisions, but also receiving this information in a clear, parsimonious and direct format that is intuitive and can be easily incorporated into the decision-making processes is key for the sustainability of their farm fields.After the operational provision of the climate forecast service, some implications of our findings have been detected: ● Combined downscaling and contextualisation. Downscaling techniques help to provide more accurate information at the location scale. Nevertheless, the spatial visualization over large regions may help to contextualize the climate forecasts, and that is also beneficial for end-user interpretation of the climate-related information. ● Skill. Efforts to provide more skilful climate forecasts are required, as well as alternative information to be used in situations where skill is lacking (e.g.: past climate information). The skill assessment is crucial to inform users about forecast reliability and accuracy. It enhances user understanding and trust in the service by generating realistic expectations. ● Operationalisation. Reducing the time for data processing or increasing the computational resources will allow the provision of climate forecasts that are closer to the date of production. Users will benefit from earlier access to the information than the current time-lapse. ● Co-production. The co-production process ensures the service is designed based on user needs and feedback which is more likely to be adopted and effectively utilized by users. This process never ends, so, it needs to be applied if future developments or requests are included in the current service. ● Support for sustainability. End-users benefit from the integration of the climate forecast along with phenological and disease risk services. Further explorations to include more climate variables in the service and in the impact models will be beneficial for end-users. For instance, the probabilities of extreme events help vineyard managers prepare for potential adverse conditions, while, safeguarding vineyards against weather impacts through proactive measures. ● Capacity building. By training users about the nature of climate forecasts, they can enhance the preparedness and resilience response to varying climate conditions. New information and products could be co-explored to co-evaluate their utility and usefulness.